Tendency Bias Correction in the GEOS AGCM - NASA/TM-2021-104606/Vol. 57 Technical Report Series on Global Modeling and Data Assimilation, Volume ...

Page created by Claude Johnston
 
CONTINUE READING
Tendency Bias Correction in the GEOS AGCM - NASA/TM-2021-104606/Vol. 57 Technical Report Series on Global Modeling and Data Assimilation, Volume ...
NASA/TM-2021-104606/Vol. 57

Technical Report Series on Global Modeling and Data Assimilation,
Volume 57

Randal D. Koster, Editor

Tendency Bias Correction in the GEOS AGCM

Yehui Chang, Siegfried Schubert, Randal Koster and Andrea Molod

August 2021
Tendency Bias Correction in the GEOS AGCM - NASA/TM-2021-104606/Vol. 57 Technical Report Series on Global Modeling and Data Assimilation, Volume ...
NASA STI Program ... in Profile

Since its founding, NASA has been dedicated  CONFERENCE PUBLICATION.
to the advancement of aeronautics and space Collected papers from scientific and
science. The NASA scientific and technical technical conferences, symposia, seminars,
information (STI) program plays a key part in or other meetings sponsored or
helping NASA maintain this important role. co-sponsored by NASA.

The NASA STI program operates under the  SPECIAL PUBLICATION. Scientific,
auspices of the Agency Chief Information Officer. technical, or historical information from
It collects, organizes, provides for archiving, and NASA programs, projects, and missions,
disseminates NASA’s STI. The NASA STI often concerned with subjects having
program provides access to the NTRS Registered substantial public interest.
and its public interface, the NASA Technical
Reports Server, thus providing one of the largest  TECHNICAL TRANSLATION.
collections of aeronautical and space science STI English-language translations of foreign
in the world. Results are published in both non- scientific and technical material pertinent to
NASA channels and by NASA in the NASA STI NASA’s mission.
Report Series, which includes the following report
types: Specialized services also include organizing
 and publishing research results, distributing
 TECHNICAL PUBLICATION. Reports of specialized research announcements and
 completed research or a major significant feeds, providing information desk and personal
 phase of research that present the results of search support, and enabling data exchange
 NASA Programs and include extensive data services.
 or theoretical analysis. Includes compila-
 tions of significant scientific and technical For more information about the NASA STI
 data and information deemed to be of program, see the following:
 continuing reference value. NASA counter-
 part of peer-reviewed formal professional  Access the NASA STI program home page
 papers but has less stringent limitations on at http://www.sti.nasa.gov
 manuscript length and extent of graphic
 presentations.  E-mail your question to help@sti.nasa.gov

 TECHNICAL MEMORANDUM.  Phone the NASA STI Information Desk at
 Scientific and technical findings that are 757-864-9658
 preliminary or of specialized interest,
 e.g., quick release reports, working  Write to:
 papers, and bibliographies that contain NASA STI Information Desk
 minimal annotation. Does not contain Mail Stop 148
 extensive analysis. NASA Langley Research Center
 Hampton, VA 23681-2199
 CONTRACTOR REPORT. Scientific and
 technical findings by NASA-sponsored
 contractors and grantees.
Tendency Bias Correction in the GEOS AGCM - NASA/TM-2021-104606/Vol. 57 Technical Report Series on Global Modeling and Data Assimilation, Volume ...
NASA/TM-2021-104606/Vol. 57

Technical Report Series on Global Modeling and Data Assimilation,
Volume 57

Randal D. Koster, Editor

Tendency Bias Correction in the GEOS AGCM

Yehui Chang
Morgan State University, Baltimore, MD

Siegfried Schubert
Science Systems and Applications Inc., Lanham, MD

Randal Koster
Global Modeling and Assimilation Office (GMAO), NASA Goddard Space Flight
Center, Greenbelt, MD

Andrea Molod
Global Modeling and Assimilation Office (GMAO), NASA Goddard Space Flight
Center, Greenbelt, MD

National Aeronautics and
Space Administration

Goddard Space Flight Center
Greenbelt, Maryland 20771

August 2021
Tendency Bias Correction in the GEOS AGCM - NASA/TM-2021-104606/Vol. 57 Technical Report Series on Global Modeling and Data Assimilation, Volume ...
Notice for Copyrighted Information
 This manuscript has been authored by employees of the National Aeronautics and
 Space Administration, Morgan State University under contract
 NNG11HP16A, Science Systems and Applications Inc. under contract NNG17HP01C,
 with the National Aeronautics and Space Administration. The United States Government
 has a non-exclusive, irrevocable, worldwide license to prepare derivative works, publish,
 or reproduce this manuscript, and allow others to do so, for United States Government
 purposes. Any publisher accepting this manuscript for publication acknowledges that the
 United States Government retains such a license in any published form of this
 manuscript. All other rights are retained by the copyright owner.

Trade names and trademarks are used in this report for identification only. Their usage
does not constitute an official endorsement, either expressed or implied, by the National
Aeronautics and Space Administration.

Level of Review: This material has been technically reviewed by technical management.

 Available from

 NASA STI Program National Technical Information
 Mail Stop 148 Service 5285 Port Royal Road
 NASA’s Langley Research Springfield, VA 22161
 Center Hampton, VA 703-605-6000
 23681-2199
Tendency Bias Correction in the GEOS AGCM - NASA/TM-2021-104606/Vol. 57 Technical Report Series on Global Modeling and Data Assimilation, Volume ...
Table of Contents

List of Tables and Figures ......................................................................................... 2

Executive Summary ................................................................................................... 7

1.0 Review of TBC and focus of this report ........................................................... 10

2.0 Experiments and datasets.................................................................................. 13

2.1 MERRA-2 and other observational datasets ................................................................................13

2.2 The GEOS AGCM .......................................................................................................................14

2.3 The TBC simulations....................................................................................................................15

3.0 Results of TBC.................................................................................................... 17

3.1 Efficacy of global and regional TBC in correcting key fields .......................................................19

 a) u-wind at 250mb ........................................................................................................................... 19
 b) Temperature at 2 meters (T2m) .................................................................................................... 25
 c) Precipitation .................................................................................................................................. 31

3.2 Corrections to other diagnostic fields ...........................................................................................38

3.3 A closer look at the boreal winter stationary waves ......................................................................46

4.0 Discussion ............................................................................................................ 53

4.1 Why TBC works and its limitations..............................................................................................53

4.2 A state-dependent extension to TBC............................................................................................58

5.0 Summary and Conclusions................................................................................ 61

Acknowledgements................................................................................................... 64

References ........................................................................................................................................65

 1
Tendency Bias Correction in the GEOS AGCM - NASA/TM-2021-104606/Vol. 57 Technical Report Series on Global Modeling and Data Assimilation, Volume ...
List of Tables and Figures

Table 1. A summary of the AGCM experiments. See Fig. 2 for definitions of the regions. The AGCM
is the MERRA-2 version of the model. For some of the results shown in Section 3.4, however, the
experiments were repeated using a more recent version of the model for the period 1981-2016 (see text
for details).

Table 2. The bias correction for the M2_AGCM for the 250mb u-wind in the NM region for TBC
applied in the NM region (second column) and the contributions from the NM2 (third column), NM4
(fourth column), and NM5 (fifth column), subregions, as well as the sum of the corrections from all
6 NM subregions (last column). The results show the decomposition of the inner product (in black,
as a percent) in terms of an amplitude ratio (in blue, as a percent) and spatial similarity (in red) as
defined by the RHS of equation 3 in the text. For the values to the left of the dashed lines, the biases
are computed with respect to MERRA-2, and the values to the right of the dashed lines are the same
except that the biases are computed with respect to ERA-5. See Fig. 2 for the definitions of the
regions.

Table 3. The bias correction for the M2_AGCM for T2m in the NM region for TBC applied in the
NM region (second column) and the contributions from the NM2 (third column), NM4 (fourth
column), and NM5 (fifth column), subregions, as well as the sum of the corrections from all 6 NM
subregions (last column). The results show the decomposition of the inner product (in black, as a
percent) in terms of an amplitude ratio (in blue, as a percent) and spatial similarity (in red) as defined
by the RHS of equation 3 in the text. For the values to the left of the dashed lines, the biases are
computed with respect to MERRA-2, and the values to the right of the dashed lines are the same
except that the biases are computed with respect to ERA-5. See Fig. 2 for the definitions of the
regions.

 2
Tendency Bias Correction in the GEOS AGCM - NASA/TM-2021-104606/Vol. 57 Technical Report Series on Global Modeling and Data Assimilation, Volume ...
Table 4. The bias correction for the M2_AGCM for precipitation in the NM region for TBC applied
in the NM region (second column) and the contributions from the NM2 (third column), NM4 (fourth
column), and NM5 (fifth column), subregions, as well as the sum of the corrections from all 6 NM
subregions (last column). The results show the decomposition of the inner product (in black, as a
percent) in terms of an amplitude ratio (in blue, as a percent) and spatial similarity (in red) as defined
by the RHS of equation 3 in the text. For the values to the left of the dashed lines, the biases are
computed with respect to MERRA-2, and the values to the right of the dashed lines are the same
except that the biases are computed with respect to ERA-5. See Fig. 2 for the definitions of the
regions

Figure 1: A schematic of the model’s climate drift. Here F is the forecast, O is the corresponding
observational value, and the overbar denotes an average over a large number of forecasts. The
example is for a positive model bias.

Figure 2: The 17 regions in which TBC was applied (see Table 1). The heavy black box outlines the
NM region (a key focus of this study).

Figure 3: Results for the 250mb u-wind for each season. The left panels are the negative of the bias
(MERRA-2 minus CNTRL) and the right panels are the impact of the TBC (TBC_GLOBAL minus
CNTRL). Values are averaged over the period 1980-2017. Units are m/s.

Figure 4: Top panels: The contributions for each season to the global TBC impacts on the 250mb u-
wind in the NM region from TBC in each of the zonal band subregions (NP, NM, TR, SM, SP). The
last bar in each plot is the sum of the contributions from each zonal band. Bottom panels: The
contributions for each season to the NM TBC impacts in the NM region from each subregion of NM
(NM1, NM2, NM3, NM4, NM5, NM6). The last bar in each plot is the sum of the contributions from
each subregion of NM. The values are the normalized inner products (I) as defined in eq. 3 of the
text. See Table 1 and Figure 2 for the definitions of the regions.

Figure 5: Results for T2m for each season. The left panels are negative of the bias (MERRA-2 -
CNTRL) and the right panels are the impact of the TBC (TBC_GLOBAL minus CNTRL). Values
are averaged over the period 1980-2017. Units are °C.

 3
Tendency Bias Correction in the GEOS AGCM - NASA/TM-2021-104606/Vol. 57 Technical Report Series on Global Modeling and Data Assimilation, Volume ...
Figure 6: Top panels: The contributions for each season to the global TBC impacts on T2m in the
NM region from TBC in each of the zonal band subregions (NP, NM, TR, SM, SP). The last bar in
each plot is the sum of the contributions from each zonal band. Bottom panels: The contributions for
each season to the NM TBC impacts in the NM region from each subregion of NM (NM1, NM2, NM3,
NM4, NM5, NM6). The last bar in each plot is the sum of the contributions from each subregion of
NM. The values are the normalized inner products as defined in eq. 3 of the text. See Figure 2 for
the definitions of the regions.

Figure 7: Results for precipitation for each season. The left panels are the negative of the bias
(MERRA-2 - CNTRL) and the right panels are the impact of the TBC (TBC_GLOBAL minus
CNTRL). Values are averaged over the period 1980-2017. Units are mm/day.

Figure 8: Top panels: The contributions for each season to the global TBC impacts on precipitation
in the NM region from TBC in each of the zonal band subregions (NP, NM, TR, SM, SP). The last
bar in each plot is the sum of the contributions from each zonal band. Bottom panels: The
contributions to the NM TBC impacts in the NM region from each subregion of NM (NM1, NM2,
NM3, NM4, NM5, NM6). The last bar in each plot is the sum of the contributions from each subregion
of NM. The values are the normalized inner products as defined in eq. 3 of the text. See Figure 2 for
the definitions of the regions.

Figure 9: T2m bias correction in NM5 from TBC in various NM Subregions. The contributions from
the NM region and from the NM2, NM4, and NM5 subregions, as well as the sum of the corrections
from all 6 NM subregions, are shown. The biases are computed with respect to station observations
(dark bars) or ERA-5 (light bars). The values are the normalized inner products as defined in eq. 3
of the text. See Figure 2 for the definitions of the regions.

Figure 10: Schematic of how the TBCs might be acting to influence the tendencies. This could be
indirectly through the impacts on the physics and dynamics terms (solid arrows), or directly (dashed
arrow).

Figure 11: Impacts for JJA of global TBC on a) total cloudiness (fraction of area), b) latent heat flux,
c) sensible heat flux, d) longwave flux at the surface and, e) shortwave flux at the surface. In each

 4
Tendency Bias Correction in the GEOS AGCM - NASA/TM-2021-104606/Vol. 57 Technical Report Series on Global Modeling and Data Assimilation, Volume ...
set of two panels the left panel is the negative of the bias (MERRA-2 minus CNTRL), and the right
panel is TBC minus CNTRL. Values are averaged over the period 1980-2017. Units are: W/m2.

Figure 12: Impacts for SON of global TBC on a) total cloudiness (fraction of area), b) latent heat
flux, c) sensible heat flux, d) longwave flux at the surface and, e) shortwave flux at the surface. In
each set of two panels the left panel is the negative of the bias (MERRA-2 minus CNTRL), and the
right panel is TBC minus CNTRL. Values are averaged over the period 1980-2017. Units are: W/m2.

Figure 13: Impacts for DJF of global TBC on a) total cloudiness (fraction of area), b) latent heat
flux, c) sensible heat flux, d) longwave flux at the surface and, e) shortwave flux at the surface. In
each set of two panels the left panel is the negative of the bias (MERRA-2 minus CNTRL), and the
right panel is TBC minus CNTRL. Values are averaged over the period 1980-2017. Units are: W/m2.

Figure 14: Impacts for MAM of global TBC on a) total cloudiness (fraction of area), b) latent heat
flux, c) sensible heat flux, d) longwave flux at the surface and, e) shortwave flux at the surface. In
each set of two panels the left panel is the negative of the bias (MERRA-2 minus CNTRL), and the
right panel is TBC minus CNTRL. Values are averaged over the period 1980-2017. Units are: W/m2.

Figure 15: Impact of TBC on the 250mb eddy (deviations from the zonal mean) height field
compared to the model bias. The left panels are the negative of the bias with respect to MERRA-2.
The right panels show the impact of TBC. The top panels are for the M2_AGCM (averaged for the
period 1980-2017). The bottom panels are for the updated model (the IC_AGCM), averaged for the
period 1981-2016. Units are meters. The box in each figure outlines our region of interest - the
Pacific/North American region.

Figure 16: A schematic of the replay approach used to compute the analysis increments (IAU, or ∆ 
in our current terminology) from an existing analysis (figure taken from Chang et al. 2019). Here
IAU refers to the incremental analysis update procedure for performing data assimilation developed
by Bloom et al. (1996). See also Takacs et al. 2018 for further information about the numerical
stability of replay.

Figure 17: The impact of global and regional TBC on the 250mb height field for the IC_AGCM. a)
negative of the bias with respect to MERRA-2, b) the impact of global TBC, c) the sum of the impacts
from the tropical (TR) and northern middle latitude (NM) TBC, d) impact of TR TBC, e) impact of

 5
Tendency Bias Correction in the GEOS AGCM - NASA/TM-2021-104606/Vol. 57 Technical Report Series on Global Modeling and Data Assimilation, Volume ...
NM TBC, f) impact of applying TBC in the region that combines TR2 and TR3 and TR4. Results are
averages for the period 1981-2016. Units are meters.

Figure 18: Same as Fig. 17, except for the precipitation. Units are mm/day.

Figure 19: Long term averages (1980-2016, denoted by an overbar) of the various terms of the JJA
mean temperature (T) tendency at 850mb computed from MERRA-2. The top 6 panels show the
contributions to the tendency from the various physical terms in the thermodynamic equation such
 ̅
that ̅̅̅̅̅̅̅̅̅̅̅̅̅ + 
 = ̅̅̅̅̅
 ̅̅̅̅̅̅̅̅ + ⋯ +△ ̅̅̅̅̅
 ≅ 0, where △ (lower middle panel) is the long-term
 
average analysis increment of temperature. The lower left panel is the sum (total) of all the physical
terms. The lower right panel shows that the long-term average of the sum of the physical terms and
the analysis increment is indeed effectively zero. Units: °C/day.

Figure 20: Long term averages (1980-2016, denoted by an overbar) of the various terms of the JJA
mean specific humidity (q) tendency at 850mb computed from MERRA-2. The top panels show the
 ̅
contributions to the tendency from the various physical terms in the moisture equation. Here, =
 
̅̅̅̅̅̅̅̅̅̅̅̅̅ + ⋯ + ̅̅̅̅̅
 + ̅̅̅̅̅̅̅̅ △ ≅ 0, where ̅̅̅̅̅
 △ (lower middle panel) is the long-term average
analysis increment of specific humidity. The lower left panel is the sum (total) of all the physical
terms. The lower right panel shows that the long-term average of the sum of the physical terms and
the analysis increment is indeed effectively zero. Units are g/day.

Figure 21: An example of the s for surface pressure for January and July.

Figure 22: A comparison of the results for TBC and TBC together with a state-dependent term
(SD_TBC) for JJA for the 850mb u-wind (m/s, left panels) and the 250mb v-wind variance ((m/s)2,
right panels).

 6
Executive Summary

The Global Modeling and Assimilation Office (GMAO) recently developed an approach for

correcting long-term biases in climate models. The approach, called tendency bias correction (TBC),

introduces additional forcing terms into the model’s prognostic equations based on the long-term

(multi-decade) averages of short term (typically 6 hour) forecast errors. The approach was further

generalized to allow limiting the TBC to specific regions of the globe, thereby providing insights

into those regions of the global where model errors have the largest impact on climate bias.

In this report, we summarize the results of global and regional TBC applied to the GEOS AGCM

(the same model used to generate MERRA-2 though run at lower resolution, referred to as the

M2_AGCM) employing the analysis increments (analysis minus forecast) generated by MERRA-2

to calculate the TBC terms. Extending the results of Chang et al. (2019) and Schubert et al. (2019),

we examine in more detail the seasonality of the TBC impacts, including a deeper look into the

reasons for why TBC appears to be least effective during boreal winter. We also look into the ability

of TBC to correct other diagnostic fields such as cloudiness and surface fluxes and, more generally,

attempt to provide some insight into why TBC is effective in correcting long term climate biases.

Looking beyond TBC, we also present some preliminary results from an extension of the TBC

approach in which we include a state-dependent term.

Focusing on the Northern Hemisphere (NH) middle latitude impacts, the results show a substantial

seasonality in the efficacy of TBC in correcting many of the long-standing circulation biases (e.g.,

upper-level jets and stationary waves) of the M2_AGCM, with the summer season showing the

greatest improvements and the winter season showing the least. A key difference between JJA and

 7
DJF is the fact that the contributions to the TBC impacts in the Northern Hemisphere middle latitudes

tend to be local in summer. That is, the sources of the TBC impacts in middle latitudes are largely

confined to the middle latitudes, while during the winter there are substantial contributions to the

impacts in the middle latitudes from other latitude bands, especially from the tropics. The transition

season impacts are somewhere in between, with SON behaving more like JJA, and MAM behaving

more like DJF. During all seasons (though less so for DJF) we see an important impact on the climate

bias in the NH middle latitudes coming from TBC in a region encompassing the Tibetan Plateau,

highlighting the importance of correcting the model biases in that very mountainous region.

Similarly, we found a substantial seasonality in the ability of TBC to correct the climate biases in the

surface meteorology over North America, with TBC correcting about 60% of the bias in T2m and

more than 50% of the bias in precipitation during JJA. In contrast, during DJF, TBC corrects only

about 30% of the bias in T2m and less than 20% of the bias in precipitation.

The impacts on other diagnostic fields (total cloudiness, latent and sensible heat fluxes and the long-

wave and short-wave radiation fluxes) also show substantial improvements, but that improvement

has a strong seasonality with the largest improvements again occurring in JJA and the smallest

improvements occurring in DJF. While we cannot rule out the possibility that the TBCs are

introducing some artificial corrections that circumvent the model physics, the improvements in the

cloudiness and surface fluxes indicate that the improvements in the climate biases from TBC is, at

least in part, the result of improved input (the prognostic quantities) to the relevant

parameterizations, leading to physically realistic improvements (via the physical parametrizations)

to such fields as T2m and precipitation.

 8
The reasons for the smaller impacts in the NH during DJF are still somewhat unclear, though a

distinguishing feature of that season is the greater importance of tropical heating errors in contributing

to NH middle latitude stationary wave biases. Using an updated GEOS AGCM (though with boreal

winter stationary wave biases that are very similar to those of the M2_AGCM) we found that the

TBCs obtained from employing a technique called “replay” developed in the GMAO to generate the

short-term forecast errors did produce improved stationary waves and related fields, and those

improvements appear to be the result of improved tropical heating with TBC. This apparent

sensitivity of the results to how the TBCs are produced (in the case of the M2_AGCM they were

simply taken from the MERRA2 archive) is of some concern, but highlights the need to compute the

short-term forecast errors (and the TBCs) for the model in question directly, in either a data

assimilation or replay environment.

We also look more generally at the reasons why TBC seems to work, looking in particular at the size

of the TBCs relative to the physical and dynamical forcing terms in the model. The results show

that the TBCs overall tend to be relatively small, suggesting the model’s response to the TBCs is

likely linear, though that is less true for the moisture in the tropics where the TBCs can have

amplitudes locally that rival those of the physical terms. We in fact do find considerable double

counting (nonlinearity) when examining the separate impacts of TBC from different regions, and

this appears to be especially true for precipitation, and for DJF during which the impacts from the

tropics is important.

Finally, we present some initial results of an extension to TBC that includes a state-dependent term.

The motivation for such an extension is that by minimizing the error in the time tendency, such a

term could potentially produce a greater positive impact (compared with the original TBC approach)

 9
on forecast skill. The results we have obtained so far highlight some of the challenges one faces in

producing statistically robust estimates of state-dependent terms, and in introducing them in a way

that maintains model stability.

1.0 Review of TBC and focus of this report

Recently Chang et al (2019) introduced the tendency bias correction (TBC) approach for correcting

biases in climate models. The TBC approach, described in more detail below, uses information about

biases in short-term (e.g., 6 hour) forecast errors to calculate constant correction terms that are

introduced into the model’s prognostic equations. Schubert et al. (2019) extended the approach to

allow limiting the TBC to specific regions (termed “regional TBC”), thereby providing a systematic

approach to identifying how model errors in any one region impact climate bias across the globe.

In this report we expand on the results of Chang et al (2019) and Schubert et al (2019) by delving

further into the ability of TBC (both global and regional) to correct the climate biases of the GEOS

AGCM, looking in more detail at the impacts during all seasons, as well as the impacts on additional

diagnostic fields (e.g., surface fluxes and cloudiness) that provide further insights into how TBC acts

to correct long term climate biases. We also look further into the reasons for the relatively poor

performance of the TBC in the boreal winter hemisphere. More generally, we discuss both why TBC

seems to work, and what may be fundamental limitations to TBC, especially regarding the

interpretation of regional TBC results. Finally, we introduce an extension of the TBC methodology

to include a state-dependent term.

 10
Our approach to TBC takes advantage of the incremental analysis update (IAU) procedure (Bloom et

al. 1996) employed in the GEOS data assimilation system (Rienecker et al. 2008) in which the

equations governing the assimilation have the form:

 = ( ) + ∆ (1),
 
where ∆ = ( − )/6ℎ is the instantaneous analysis increment (applied to the

model’s prognostic variables: temperature, winds, humidity, surface pressure and ozone), and f(x)

consists of all the dynamics and physics terms of the model (basically the uncorrected model).

Following Chang et al. (2019), the governing equations for the TBC approach have the same form as

(1), except that now the forcing term is a long term mean of the increments. In particular,

 ̅̅̅̅
 = ( ) + ∆ (2),
 
 ̅̅̅̅ is the time average (denoted by the overbar) of the instantaneous ∆ values computed
where ∆ 

 ̅̅̅̅ is the tendency bias correction (TBC) term. The
here over the years 1980 - 2017. In the above, ∆ 

averaging is done independently for different times of the day and for different times of the year; in

this way, we retain the diurnal and annual cycles in the TBCs. It is important to emphasize that the

TBC approach only makes use of information about the initial short-term forecast errors and

furthermore only the long-term average of those (tendency) errors. As such, it is not at all clear that

correcting tendency biases should have a substantial impact on the model’s long-term climate biases

(Figure 1). In fact, as we shall see, TBC does correct long term climate biases, and we attempt to

address why that is the case later in this report. It is also worth noting that TBC appears to provide

only modest improvements to forecast skill (Chang et al. 2019) – something we address in this report

in terms of a potential extension of TBC that more directly addresses tendency errors.

 11
In the following, we will look in detail at the impacts of TBC (applied both globally and regionally)

on the GEOS AGCM’s long term climate biases. Section 2 describes the verification data sets, the

GEOS AGCM, and the various model experiments. The results are presented in Section 3. In Section

4 we discuss the advantages and limitations of TBC as well as a possible extension to TBC involving

a state-dependent correction term. The summary and conclusions are given in Section 5.

 Figure 1: A schematic of the model’s climate drift. Here F is the forecast, O is the
 corresponding observational value, and the overbar denotes an average over a large number
 of forecasts. The example is for a positive model bias.

 12
2. Experiments and Datasets

2.1 MERRA-2 and other observational datasets

The atmospheric reanalysis data used for this study is the Modern-Era Retrospective analysis for

Research and Applications version 2 (MERRA-2; Gelaro et al. 2017). MERRA-2, developed by

NASA Goddard Space Flight Center (GSFC) / Global Modeling and Assimilation Office (GMAO),

is an updated version of MERRA (Rienecker et al. 2011) including an improvement of the

assimilating model’s physical parameterizations of moist processes, turbulence, land and ocean

surface processes, and gravity wave drag (Bosilovich et al. 2015; Molod et al. 2015; Gelaro et al.

2017; see also below). Other differences from MERRA include aerosol data assimilation, as well as

new developments in the representation of ozone and the use of precipitation observations to force

the land surface. The horizontal resolution of the MERRA-2 data is 0.625° longitude × 0.5° latitude.

The key variables used here consist of 2-meter air temperature (T2M), precipitation (the raw values

from model, i.e., the values not yet corrected by observations), zonal wind, and geopotential height.

We note that the MERRA-2 precipitation used in this study for verification is an observationally-

corrected product in which the precipitation generated by the atmospheric model underlying

MERRA-2 was merged with gauge and satellite precipitation observations (Reichle and Liu 2014,

Reichle et al. 2017). In the following, we will use the words observations and MERRA-2

interchangeably with, of course, the understanding that MERRA-2 is a reanalysis product that

combines a model-based first guess with observations and, as such, the reanalysis products are

potentially impacted by model biases.

 13
While most of our validation of the TBC results is done against MERRA-2, we also, on a more limited

basis, validate against ERA-5 (C3S 2017), and two station-based observational datasets: the gridded

GHCN_CAMS data for T2m over land (Fan and van den Dool 2008), and the GPCP V2.3 for

precipitation (Adler et al. 2003).

2.2 The GEOS AGCM

Most of the results presented here are based on the same version of the GEOS AGCM that was used

to generate MERRA-2, though run here at a lower horizontal resolution (approximately 1°); we refer

to this as the M2_AGCM. As described in Gelaro et al. (2017), M2_ AGCM includes the finite-

volume dynamical core of Putman and Lin (2007), which uses a cubed sphere horizontal

discretization at an approximate resolution of 0.5° × 0.625°, with 72 hybrid-eta levels from the

surface to 0.01 hPa. Recent upgrades to the physical parameterization schemes include increased

re-evaporation of frozen precipitation and cloud condensate, changes to the background gravity

wave drag, and an improved relationship between the ocean surface roughness and ocean surface

stress (Molod et al. 2015). The model also includes a Tokioka-type trigger on deep convection as

part of the Relaxed Arakawa-Schubert (RAS, Moorthi and Suarez 1992) convective

parameterization scheme, which governs the lower limit on the allowable entrainment plumes

(Bacmeister and Stephens 2011). The implementation of new glaciated land representation and

seasonally-varying sea ice albedo produced improved air temperatures and reduced biases in the net

energy flux over these surfaces (Cullather et al. 2014). The model includes the Catchment land

surface model developed by Koster et al. (2000). Further details about the GEOS AGCM can be

found in Molod et al. (2015).

 14
While most of the results presented here are based on the M2_AGCM described above, a few results

(presented in Section 3.3) are based on a newer version of the GEOS AGCM. This interim

development version of the model (Icarus-3_3_p2, referred to hereafter as the IC_AGCM) was also

run at 1° horizontal resolution.

2.3 The TBC Simulations

As mentioned above, all of the simulations described here (with the exception of the results shown

in Section 3.3) were produced with the same version of the GEOS AGCM that was used to produce

MERRA-2 (the M2_AGCM). The runs were made with the same SST data, greenhouse gases

(GHGs), and other forcing used to produce MERRA-2. The only differences are that the M2_AGCM

was run at coarser resolution and, of course, that the simulations did not assimilate observations.

This similarity offers the unique opportunity to assess how the observations influence various

aspects of the model climate, but of course to the extent that any model errors are reflected in the

reanalysis, it also has the potential to bias our assessment of model errors. We address the latter in

a limited way, by computing the model’s climate biases using several different verification datasets.

The main set of experiments with the M2_AGCM examined here are identical to those described in

Schubert et al. (2019) and are, for convenience, listed in Table 1. All the runs are forced with

observed SST and span the period 1980-2017. These include a control run without TBC applied

(CNTL), a run in which the TBC is applied globally (TBC_GLOBAL), and 17 runs in which the TBC

is applied over specified regions (see Figure 2). The 17 regions consist of 5 zonal bands (the Northern

Polar region or NP, the Northern Middle latitudes or NM, the Tropical Region or TR, the Southern

 15
Middle latitudes or SM, and the Southern Polar region or SP). The NM and TR latitudinal bands are

each split into six sub-regions. As mentioned above, the runs listed in Table 1 were repeated with an

updated version of the GEOS-AGCM (IC_AGCM) for a slightly shorter time period (1981-2016). In

addition, we carried out an experiment with the IC-AGCM in which TBC was applied to a larger

tropical subregion spanning much of the Indian and Pacific Oceans (combining regions TR2, TR3 and

TR4) as described in Section 3.3.

 Table 1. A summary of the GEOS AGCM experiments. See Fig. 2 for definitions of the
 regions. The AGCM is the MERRA-2 version of the model (M2_AGCM). For some of the
 results shown in Section 3.3, however, the experiments were repeated using a more recent
 version of the model (IC_AGCM) for the period 1981-2016 (see text for details).
 Exp.
 Exp. Name Description Model
 #
 1 37- year control simulation Uncorrected
 CNTL
 for the period 1980-2017 AGCM
 2 TBC_NP
 37- year simulations for the
 TBC_NM
 period 1980-2017 in which
 TBC_TR AGCM with TBC
 TBC is applied to selected
 TBC_SM
 zonal bands
 TBC_SP
 3 TBC_NM1
 37- year simulations for the
 TBC_NM2
 period 1980-2017 in which
 TBC_NM3
 TBC applied to selected AGCM with TBC
 TBC_NM4
 regions spanning the NM
 TBC_NM5
 region
 TBC_NM6
 4 TBC_TR1
 37- year simulations for the
 TBC_TR2
 period 1980-2017 in which
 TBC_TR3
 TBC is applied to selected AGCM with TBC
 TBC_TR4
 regions spanning the TR
 TBC_TR5
 region
 TBC_TR6
 5 37- year simulations for the
 TBC_GLOBAL period 1980-2017 in which AGCM with TBC
 TBC is applied globally

 16
Figure 2: The 17 regions in which TBC was applied (see Table 1). The heavy black box
 outlines the NM region (a key focus of this study).

3. Results of TBC

In assessing the efficacy of the TBC approach, we distinguish in the following between “impact”

and “bias correction”, with the former being an assessment of how TBC influences the model’s

climate, and the latter being an assessment of the extent to which TBC actually corrects the model’s

climate bias (i.e., to what extent do the impacts of TBC project onto the climate bias). These are

quantified with an inner product measure (Schubert et al. 2019):

 • | |
 I= | |2
 = | | , (3)

 17
where X and Y are vectors with components and , respectively, at the grid points (j) making up

a particular region. Here X and Y can refer to the results (TBC minus control) of two different TBC

runs (when addressing impacts) or, in the case of assessing climate bias correction, Y refers to the

climate bias (control minus observed)1. The former (addressing impacts) is especially important for

quantifying the contributions made to the impact of TBC applied in one region (e.g., the NM region,

see Figure 2), from TBC applied to various subregions (e.g., NM1, NM2 ..., etc.), thereby allowing

us to potentially identify those subregions having the largest impacts. The right-hand side of (3)

indicates that the inner product between X and Y can be written as the ratio of the magnitudes of the

vectors, multiplied by the cosine of the angle ( ) between the two vectors (a measure of spatial

similarity).

We shall see that regional TBC does not always produce results that are easy to interpret. Specifically,

the results are in some cases (especially for precipitation) not linear in the sense that the sum of the

responses (Ri, i=1,N) obtained when TBC is applied separately to each of the subregions making up

a larger region add up to a value that is larger than the response (R) obtained when TBC is applied to

that larger region. That is,

 R1 + R2 + … RN > R (4).

 1
 Here we actually compute (observed minus control) for the sign to be consistent with how we
compute the impacts (TBC minus control).
 18
This nonlinearity (which we refer to as double counting) appears to reflect the fact that improvements

in one region appear to be coming from TBC applied to two (or more) different regions for which

the tendency errors are not independent.

We begin in Section 3.1 by examining the extent to which global and regional TBC corrects the

climates of the 250mb zonal wind, the 2-meter temperature (T2m) over land, and the precipitation.

This includes an assessment of both how TBC applied to one region impacts the climate in other

regions and the extent to which those impacts act to reduce the M2_AGCM’s climate bias.

3.1 Efficacy of global and regional TBC in correcting key fields

 a) u-wind at 250mb

Figure 3 shows the results of applying TBC globally for the 250mb u-wind for each season, with the

bias (actually the negative of the bias with respect to MERRA-2) shown in the left panels and the

TBC impacts (TBC minus CNTRL) shown in the right panels. Comparing the anomalies in the left

and right panels, one finds a remarkable degree of similarity for all seasons. This is especially true

for JJA (cf. Figs. 3e and f) where the large summer jet biases are substantially reduced. In fact, the

percent of the 250mb u-wind bias corrected by global TBC (I, expressed as a percent) is 65%, 60%,

74%, and 58% for DJF, MAM, JJA and SON, respectively (see values in the right margin Fig. 3).

The pattern similarity ( ) is also substantial, with values ranging from 0.65 in SON to as high as

0.87 in JJA. The ratios of the magnitudes of the vectors (see 3) also show that TBC_GLOBAL

corrects a substantial fraction of the bias with values ranging from a low of 0.79 for MAM to a high

of 0.89 for SON. One result not reflected in the global inner products (but see the discussion of

 19
Table 2 below), is the lack of a substantial correction to the DJF North Pacific jet bias (cf. Figs. 3a

and b). We will look into the possible reasons for that in Section 3.3.

We next look in more detail at the nature of the TBC impacts on the 250mb zonal wind, focusing on

the impacts in the NH middle latitudes (region NM, see Fig. 2). The top panels of Figure 4 show

how much TBC, when applied separately to each of the different zonal bands, contributes to the

impacts in the NM zonal band produced with global TBC. For each season it is clear that the TBC

applied to the NM region is the main contributor to the impacts in that band. In the case of JJA (Fig.

4c), the TBC applied to the NM region accounts for essentially all the impact in that zonal band.

Also, for both MAM and SON (Figs. 4b and d), TBC in the NM region accounts for more than 80%

of the total impacts in the NM region. It is only for DJF (Fig. 4a) that other zonal bands contribute

substantially to the NM impacts, with TBC from NP and TR together accounting for about 40% of

the impact compared with 60% for the NM region. It is noteworthy that the SUM is greater than 1

(especially for SON, Fig. 4d), indicating that there is some double counting. We will come back to

the double counting issue in our discussion of the limitations of regional TBC in Section 4.1.

Given the importance of TBC in the NM region, we next examine which subregions of NM seem to

play an important role. To address that, we show in the bottom row of Figure 4 the contributions of

the TBC applied to each of the six NM subregions to the total impact from TBC in the NM region.

A key result is that the NM2 region is an important contributor during all seasons, contributing

between 40% and 60% of the total depending on the season. The reasons for the importance of the

NM2 region during JJA, as already discussed in Schubert et al. (2019), have to do with the substantial

temperature biases of the M2_AGCM over and around the Tibet highlands. We now see that the

 20
(c) (d)

 (e) (f)

 (g) (h)

Figure 3: Results for the 250mb u-wind for each season. The left panels are the negative of
the bias (MERRA-2 minus CNTRL) and the right panels are the impact of the TBC
(TBC_GLOBAL minus CNTRL). Values are averaged over the period 1980-2017. Units
are m/s.

 21
Figure 4: Top panels: The contributions for each season to the global TBC impacts on the
 250mb u-wind in the NM region from TBC in each of the zonal band subregions (NP, NM,
 TR, SM, SP). The last bar in each plot is the sum of the contributions from each zonal band.
 Bottom panels: The contributions for each season to the NM TBC impacts in the NM region
 from each subregion of NM (NM1, NM2, NM3, NM4, NM5, NM6). The last bar in each plot
 is the sum of the contributions from each subregion of NM. The values are the normalized
 inner products (I) as defined in eq. 3 of the text. See Table 1 and Figure 2 for the definitions
 of the regions.

NM2 region is also important for the other seasons; it is, in fact, even more important for both MAM

(Fig. 4f) and SON (Fig. 4h), for which the NM2 region accounts for nearly 60% of the total impact

in the NM region, compared with about 40% for JJA.

 22
Table 2 summarizes the u250mb u-wind climate bias corrections in the NM zonal band in terms of

the inner product, the amplitude ratio and similarity (eq. 3). For each result, two numbers are shown

(separated by dashed lines) corresponding to two different validation products used to compute the

bias (MERRA-2 on the left and ERA-5 on the right). Overall, the results show that there is little

difference between the two estimates of the bias for the 250mb u-wind. The percent of the u250mb

wind bias in the NM region corrected by applying TBC to the NM region is shown in the second

column of Table 2. The results show a strong seasonality to the bias correction in the NM region,

with close to 90% corrected in JJA, about 70% during MAM and SON, and only about 20% during

DJF. This is reflected in the spatial similarity which shows values of 0.91 for JJA, about 0.7 for

MAM and SON, but only about 0.3 for DJF. The amplitudes of the corrections are reasonable (close

to 100%) during all seasons except DJF, when the amplitude of the impact is only ¾ of the climate

bias. The third, fourth and fifth columns of Table 2 show the contributions to the corrections in the

NM region from TBC applied to regions NM2, NM4 and NM5, respectively. These results are

consistent with what we already saw for the impacts (Fig 4), with region NM2 playing a key role. In

particular, we see that TBC in NM2 accounts for more than 50% of the bias correction in the NM

region for MAM, about 45% for SON, 36% for JJA, but only about 20% for DJF. It is interesting

that the spatial similarity from TBC in the NM2 region is about 0.6 for all seasons except for DJF,

when it is just over 0.4 (third column of Table 2). TBC applied to the other two regions (NM 4 and

NM5) account for considerably less of the bias correction in the NM region, though both JJA and

SON do show values ranging from about 13% to 19%. It is noteworthy that both SON and MAM

show considerable double counting, with the SUM (inner product of the sum of the corrections over

all 6 subregions) for each exceeding 100%; the SUM is 115% and about 125% for SON and MAM,

respectively, reflecting excessive amplitude ratios that exceed 200% in the case of MAM.

 23
Table 2. The bias correction for the M2_AGCM for the 250mb u-wind in the NM region for TBC
applied in the NM region (second column) and the contributions from the NM2 (third column), NM4
(fourth column), and NM5 (fifth column) subregions, as well as the sum of the corrections from all 6
NM subregions (last column). The results show the decomposition of the inner product (in black, as
a percent) in terms of an amplitude ratio (in blue, as a percent) and spatial similarity (in red) as defined
by the RHS of equation 3 in the text. For the values to the left of the dashed lines, the biases are
computed with respect to MERRA-2, and the values to the right of the dashed lines are the same except
that the biases are computed with respect to ERA-5. See Fig. 2 for the definitions of the regions.

 24
b) Temperature at 2 meters (T2m)

The results of TBC for T2m are shown in Fig. 5. Recall that the AGCM is forced with observed

SSTs so we only show the results over land areas. The bias in T2m (left panels of Figure 5) has a

strong seasonality, with DJF (Fig. 5a) having a substantial cold bias over northern Eurasia, Canada

and parts of the U.S. (recall the figures show the negative of the bias), and JJA (Fig. 5e) displaying

a strong warm bias in those same regions. The biases of the transition seasons fall in between the

two extreme seasons, with MAM (Fig. 5c) displaying some of the cold bias of DJF, and SON (Fig.

5g) showing some of the warm bias of JJA. The global TBC acts to substantially correct the T2m

biases (right panels of Fig. 5), though not to same extent that we saw previously for the 250mb u-

wind. Here, the percent of the global T2m bias that is corrected is 40%, 51%, 60%, and 35%, for

DJF, MAM, JJA and SON, respectively (see values in the right margin of Fig. 5). We note that the

relatively low value of the inner product (I) in DJF (cf. Figs. 5a and b) is due to the weak amplitude

(52%) of the correction rather than a low spatial similarity (it is relatively high at 0.77). In contrast,

the relatively low value of the inner product (I) in SON (cf. Figs. 5g and h), is due to a rather low

spatial similarity of the correction (0.45) rather than a weak amplitude (it has the relatively large

value of 72%).

We next look in more detail at the nature of the T2m impacts, focusing again on the impacts in the

NH middle latitude regions (region NM, see Fig. 2). The top panels of Figure 6 show how much

TBC, when applied separately to each of the different zonal bands, contributes to the impacts seen

in the NM zonal band from global TBC. The results are quite similar to what we saw for the 250mb

 25
Figure 5: Results for T2m for each season. The left panels are negative of the bias
(MERRA-2 - CNTRL) and the right panels are the impact of the TBC (TBC_GLOBAL
minus CNTRL). Values are averaged over the period 1980-2017. Units are °C.

 26
u-wind, in that most of the correction to T2m that occurs in the NM zonal band (again, only over the

land areas) is due to the TBC applied just in the NM zonal band. The bottom row of Figure 6 shows

the contributions of the TBC applied to each of the six NM subregions to the impact seen over the

NM5 (North America) region from TBC in the NM region. While the NM2 region is again an

important contributor during JJA (Fig. 6g) and SON (Fig. 6h), contributing roughly 30% of the total

impact, that is not the case for DJF (Fig. 6e) and MAM (Fig. 6f). In fact, during DJF the main

contributor to the impact over the NM5 region is local (the TBC applied to the NM5 region alone).

During MAM, both the upstream (NM4) and local (NM5) regions play a role, while for both JJA and

SON the three regions NM2, NM4 and NM5 all roughly contribute the same amount (one third) to

the impacts over the NM5 region.

Turning to Table 3, we examine in more detail the T2m bias correction that occurs over North

America (the NM5 region). Here again, two numbers are shown (separated by dashed lines)

corresponding to two different validation products used to compute the bias (MERRA-2 on the left

and ERA-5 on the right). Focusing on the second column of Table 3, we see that the percent of the

bias in region NM5 corrected by TBC applied to the NM zonal band accounts for about 62% for JJA,

43% for SON, 29% for DJF, and 26% for MAM, with some mostly minor differences between the

results for the two different validation products. In the above, we have averaged the two values

computed from the two different validation products. The relatively low percent of bias correction

during DJF is largely a reflection of a weak amplitude of the correction (the spatial similarity is

0.86). In contrast, the relatively small bias correction for MAM is more a reflection of a relatively

weak spatial similarity, equal to 0.57. It is noteworthy that the percent of the T2m bias in the NM

region corrected by TBC is, for all seasons, slightly larger when using ERA5 for validation. Why

 27
Figure 6: Top panels: The contributions for each season to the global TBC impacts on T2m
 in the NM region from TBC in each of the zonal band subregions (NP, NM, TR, SM, SP).
 The last bar in each plot is the sum of the contributions from each zonal band. Bottom panels:
 The contributions for each season to the NM TBC impacts in the NM5 (North American)
 region from each subregion of NM (NM1, NM2, NM3, NM4, NM5, NM6). The last bar in each
 plot is the sum of the contributions from each subregion of NM. The values are the normalized
 inner products as defined in eq. 3 of the text. See Figure 2 for the definitions of the regions.

that should be the case is unclear. Overall, the results indicate that TBC appears to be more effective

in reducing the T2m bias during the summer and fall seasons, and less effective during the winter

and spring seasons. The third, fourth and fifth columns of Table 3 show the contributions to the

corrections in the NM5 region from TBC applied to regions NM2, NM4 and NM5, respectively. The

 28
results highlight the importance of the TBC applied locally during all seasons (accounting for

between 20% - 25% of the correction in each season). In fact, during DJF and MAM it is all local

(NM5), consistent with what we saw for the DJF impacts (Fig. 6e). This is not the case for MAM

which, while showing some impacts on the NM5 region from the TBC in the NM4 region (Fig. 6f),

produces essentially no bias correction from that region (Table 3, bottom cell of fourth column).

This highlights the fact that impacts do not necessarily translate into actual bias corrections. Both

JJA and SON have contributions from all three regions, with JJA having roughly three equal

contributions, and SON having NM2 and NM4 contributions that are roughly half of the local

contribution. As such, for JJA the corrections to the bias in the NM5 region are roughly 2/3 remote

and 1/3 local, while for SON, it is roughly ½ remote and ½ local.

 29
Table 3. The bias correction for the M2_AGCM for T2m in the NM5 (North American) region for
TBC applied in the NM region (second column) and the contributions from the NM2 (third
column), NM4 (fourth column), and NM5 (fifth column) subregions, as well as the sum of the
corrections from all 6 NM subregions (last column). The results show the decomposition of the
inner product (in black, as a percent) in terms of an amplitude ratio (in blue, as a percent) and
spatial similarity (in red) as defined by the RHS of equation 3 in the text. For the values to the left
of the dashed lines, the biases are computed with respect to MERRA-2, and the values to the right
of the dashed lines are the same except that the biases are computed with respect to ERA-5. See
Fig. 2 for the definitions of the regions.

 30
c) Precipitation

Figure 7 is the same as Figure 3, but for precipitation. The model has rather substantial precipitation

biases throughout the tropics during all seasons (left panels of Fig. 7). During JJA (Fig. 7e),

excessive precipitation extends from the central tropical Pacific into the northern subtropical Pacific,

while there is too little precipitation over the warm pool region and over the eastern tropical Indian

Ocean. The DJF (Fig. 7a) tropical biases have a north/south dipole structure with too much

precipitation north of the equator and too little south of the equator, especially over the Indian Ocean

region. Other biases prevalent during all seasons include excessive precipitation over southeast

Asia (though less so for DJF, Fig. 7a), northern South America extending south over the Andes

Mountains, and much of Central America and surrounding regions, extending into the Atlantic. The

model also produces too little precipitation over the central U.S. during the summer (Fig. 7e), and

too little precipitation in the North Pacific storm tracks during all but the winter season (though there

is a precipitation deficit in the northeast Pacific during that season, Fig. 7a). Global TBC acts to

correct some of these biases, though not to same extent that we saw previously for the 250mb u-

wind or even for T2m. Here the percent of the global precipitation bias that is corrected with global

TBC is 27%, 24%, 30%, and 22%, for DJF, MAM, JJA and SON, respectively (see values in the

right margin of Fig. 7). The spatial similarity is generally low, ranging from 0.36 in MAM to 0.50

for JJA. The amplitude ratios are 68%, 67%, 60%, and 57%, for DJF, MAM, JJA and SON,

respectively. As such, the overall low values of the percent precipitation bias corrections reflect

both a smaller spatial similarity and a weaker amplitude ratio compared to the results for T2m and

250mb u-wind.

 31
Turning next to Figure 8, we show how much the TBC applied separately to each of the different

zonal bands contributes to the precipitation impacts in the NM zonal band from global TBC. The

results are again quite similar to what we saw for the 250mb u-wind, in that most of the correction

to precipitation that occurs in the NM zonal band is due to the TBC applied just in the NM zonal

band. Only DJF (Fig. 8a) and SON (Fig. 8d) show some contribution from the tropics (greater than

0.2). There is also considerable double counting especially for MAM (Fig. 8b) and SON (Fig. 8d),

for which the SUM is substantially larger than 1 (values are about 1.4 for both). The bottom row

of Figure 8 shows the contributions of the TBC applied to each of the six NM subregions to the

impact on precipitation over the NM5 (North America) region from TBC in the NM region. Here

we see less of a predominant impact of the NM2 region, with the other regions (especially the just-

upstream NM4 and local NM5 regions) contributing as much if not more to the impact over North

America. Again, DJF (Fig. 8e) is somewhat of an outlier, with the largest contributions coming from

the local (NM5) and downstream (NM6) regions. Here again there is also considerable double

counting for all but JJA (Fig. 8g), especially for DJF for which the SUM is greater than 2.

Turning now to Table 4, we examine in more detail the precipitation bias correction that occurs over

North America (the NM5 region). Here again, two numbers are shown (separated by a dashed line)

corresponding to two different validation products used to compute the bias (MERRA-2 on the left

and ERA-5 on the right). Focusing on the second column of Table 4, we see that the percent of the

bias in region NM5 corrected by TBC (I, the values in black) applied to the NM zonal band accounts

for about 55% for JJA, 38% for SON, 15% for DJF, and 23% for MAM (here again we are averaging

the two values computed from the two different validation products). This highlights the large

 32
Figure 7: Results for precipitation for each season. The left panels are the negative of the
bias (MERRA-2 - CNTRL) and the right panels are the impact of the TBC (TBC_GLOBAL
minus CNTRL). Values are averaged over the period 1980-2017. Units are mm/day.

 33
Figure 8: Top panels: The contributions for each season to the global TBC impacts on
 precipitation in the NM region from TBC in each of the zonal band subregions (NP, NM,
 TR, SM, SP). The last bar in each plot is the sum of the contributions from each zonal band.
 Bottom panels: The contributions to the NM TBC impacts in the NM5 (North American)
 region from each subregion of NM (NM1, NM2, NM3, NM4, NM5, NM6). The last bar in each
 plot is the sum of the contributions from each subregion of NM. The values are the
 normalized inner products as defined in eq. 3 of the text. See Figure 2 for the definitions of
 the regions.

seasonal differences in the corrections, with JJA exhibiting the largest bias correction, and DJF

exhibiting the smallest. We note that in this case there are more substantial differences between the

results for the two different validation products (MERRA-2 and ERA-5) compared with the winds

and temperature, reflecting the greater uncertainties in our observational estimates of precipitation,

 34
though again the percent corrected is larger when using ERA-5 as validation. Turning to columns

3, 4 and 5 of Table 4, we see that both JJA and SON exhibit relatively large local contributions

(NM5, about 25%), with JJA also showing a relatively large contribution from NM2 (more than

21%), while SON also shows a relatively large contribution from the upstream region (NM4, about

23%). For all but JJA, there is again considerable double counting as reflected in the SUM (Last

column of Table 4).

The T2m and precipitation bias corrections over North America (the NM5 region) are summarized

in Figure 9 for JJA and DJF – the two seasons that show the largest differences in the amount of bias

correction obtained from TBC. The results are shown for the impact of the full NM region (left

most bars), the NM2, NM4 and NM5 regions, and the SUM of all 6 regions (right most bars). Here

the two bars for each region again correspond to two different validation datasets, one being ERA-

5 (for both T2m and precipitation) and the other being the GHCN_CAMS station dataset (for T2m)

or the GPCP V2.3 station dataset (for precipitation). For T2m over North America (top panels of

Fig. 9), the results highlight the substantial differences between JJA (Fig. 9a) and DJF (Fig. 9b) in

terms of the percent corrected as well as the type of correction, with about double the total correction

in JJA (60% for JJA versus 30% for DJF), and with 2/3 of the JJA correction coming from remote

regions (NM2 and NM4) and 1/3 local (NM5). In contrast, during DJF essentially the entire

correction is local (NM5). For precipitation (bottom panels of Fig. 9), the results again highlight the

substantial differences between JJA (Fig. 9c) and DJF (Fig. 9d) in terms of the percent corrected as

well as the type of correction, with almost a factor of 4 larger total correction in JJA (55% for JJA

versus 15% for DJF), and with 1/2 of the JJA correction coming from remote regions (NM2 and

NM4) and 1/2 local (NM5). Interestingly, none of the precipitation bias correction during DJF is

 35
You can also read